Make nvme_ctrlr_reset a nop if a reset is already in progress.

This protects against cases where a controller crashes with multiple
I/O outstanding, each timing out and requesting controller resets
simultaneously.

While here, remove a debugging printf from a previous commit, and add
more logging around I/O that need to be resubmitted after a controller
reset.

Sponsored by:	Intel
Reviewed by:	carl

3 files changed, 29 insertions, 5 deletions

diff --git a/sys/dev/nvme/nvme_ctrlr.c b/sys/dev/nvme/nvme_ctrlr.c
index cf6889e..762cfe6 100644
--- a/sys/dev/nvme/nvme_ctrlr.c
+++ b/sys/dev/nvme/nvme_ctrlr.c
@@ -422,6 +422,13 @@ nvme_ctrlr_hw_reset(struct nvme_controller *ctrlr)
 void
 nvme_ctrlr_reset(struct nvme_controller *ctrlr)
 {
+	int cmpset;
+
+	cmpset = atomic_cmpset_32(&ctrlr->is_resetting, 0, 1);
+
+	if (cmpset == 0)
+		/* Controller is already resetting. */
+		return;
 
 	taskqueue_enqueue(ctrlr->taskqueue, &ctrlr->reset_task);
 }
@@ -700,6 +707,8 @@ nvme_ctrlr_reset_task(void *arg, int pending)
 	pause("nvmereset", hz / 10);
 	if (status == 0)
 		nvme_ctrlr_start(ctrlr);
+
+	atomic_cmpset_32(&ctrlr->is_resetting, 1, 0);
 }
 
 static void
@@ -896,6 +905,8 @@ intx:
 	    taskqueue_thread_enqueue, &ctrlr->taskqueue);
 	taskqueue_start_threads(&ctrlr->taskqueue, 1, PI_DISK, "nvme taskq");
 
+	ctrlr->is_resetting = 0;
+
 	return (0);
 }
 
diff --git a/sys/dev/nvme/nvme_private.h b/sys/dev/nvme/nvme_private.h
index bae9aaa..406a2e0 100644
--- a/sys/dev/nvme/nvme_private.h
+++ b/sys/dev/nvme/nvme_private.h
@@ -100,7 +100,7 @@ MALLOC_DECLARE(M_NVME);
 #define NVME_MAX_CONSUMERS	(2)
 #define NVME_MAX_ASYNC_EVENTS	(8)
 
-#define NVME_DEFAULT_TIMEOUT_PERIOD	(30)	/* in seconds */
+#define NVME_DEFAULT_TIMEOUT_PERIOD	(30)    /* in seconds */
 #define NVME_MIN_TIMEOUT_PERIOD		(5)
 #define NVME_MAX_TIMEOUT_PERIOD		(120)
 
@@ -280,6 +280,8 @@ struct nvme_controller {
 
 	void				*cons_cookie[NVME_MAX_CONSUMERS];
 
+	uint32_t		is_resetting;
+
 #ifdef CHATHAM2
 	uint64_t		chatham_size;
 	uint64_t		chatham_lbas;
diff --git a/sys/dev/nvme/nvme_qpair.c b/sys/dev/nvme/nvme_qpair.c
index 602171f..b6a2633 100644
--- a/sys/dev/nvme/nvme_qpair.c
+++ b/sys/dev/nvme/nvme_qpair.c
@@ -142,7 +142,13 @@ nvme_qpair_complete_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr,
 		TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq);
 		TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq);
 
-		if (!STAILQ_EMPTY(&qpair->queued_req)) {
+		/*
+		 * If the controller is in the middle of resetting, don't
+		 *  try to submit queued requests here - let the reset logic
+		 *  handle that instead.
+		 */
+		if (!STAILQ_EMPTY(&qpair->queued_req) &&
+		    !qpair->ctrlr->is_resetting) {
 			req = STAILQ_FIRST(&qpair->queued_req);
 			STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq);
 			_nvme_qpair_submit_request(qpair, req);
@@ -462,8 +468,6 @@ nvme_timeout(void *arg)
 
 	/* Read csts to get value of cfs - controller fatal status. */
 	csts.raw = nvme_mmio_read_4(ctrlr, csts);
-	device_printf(ctrlr->dev, "i/o timeout, csts.cfs=%d\n", csts.bits.cfs);
-	nvme_dump_command(&tr->req->cmd);
 
 	if (ctrlr->enable_aborts && csts.bits.cfs == 0) {
 		/*
@@ -606,8 +610,12 @@ nvme_io_qpair_enable(struct nvme_qpair *qpair)
 
 	nvme_qpair_enable(qpair);
 
-	TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq)
+	TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq) {
+		device_printf(qpair->ctrlr->dev,
+		    "resubmitting outstanding i/o\n");
+		nvme_dump_command(&tr->req->cmd);
 		nvme_qpair_submit_tracker(qpair, tr);
+	}
 
 	STAILQ_INIT(&temp);
 	STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request);
@@ -615,6 +623,9 @@ nvme_io_qpair_enable(struct nvme_qpair *qpair)
 	while (!STAILQ_EMPTY(&temp)) {
 		req = STAILQ_FIRST(&temp);
 		STAILQ_REMOVE_HEAD(&temp, stailq);
+		device_printf(qpair->ctrlr->dev,
+		    "resubmitting queued i/o\n");
+		nvme_dump_command(&req->cmd);
 		_nvme_qpair_submit_request(qpair, req);
 	}